Artificial intelligence has entered a transformative era with the advent of large foundation models, including large language models (LLMs), large vision models (LVMs), and large multimodal models (LMMs). These models have demonstrated exceptional capabilities in enhancing natural language understanding, computer vision, and multimodal tasks. Radiology, a field facing increasing demands due to the growing volume and complexity of medical imaging, stands to benefit significantly from these advancements. Foundation models offer the possibility to aid radiologists in triaging cases, generating accurate reports, and identifying potential abnormalities. Establishing effective, radiology-specific models could revolutionize diagnostic processes and improve overall clinical outcomes.

This Research Topic aims to facilitate the development and application of foundation models in the radiological domain. The goal is to present innovative research exploring the adaptation and implementation of LLMs, LVMs, and LMMs in radiology. By focusing on the specific challenges inherent to radiology, such as the sheer volume of imaging data and task complexity, this research topic seeks to advance our understanding of how these models can enhance clinical workflows, improve diagnostic precision, and ultimately elevate patient care standards. Contributions that display the potential of these technologies to successfully integrate into radiological practices are especially encouraged.

To gather further insights into the integration and impact of foundation models in radiology, we welcome articles addressing, but not limited to, the following themes:

• New neural network architecture design for radiology-specific foundation models.
• Training, fine-tuning, and adaptation methods tailored for radiology.
• Algorithms for multimodal alignment within radiology-specific models.
• Paradigms for incorporating human feedback into model training.
• Optimization, acceleration, and compression of radiology-specific models.
• Frameworks for integrating multiple models in radiology applications.
• Privacy-preserving training techniques for radiology models.
• Interpretability and explainability considerations for foundation models.
• Trustworthiness, including ethical considerations and bias mitigation.
• Evaluation metrics, open datasets, and benchmarking for radiological applications.

In addition to original research articles, we invite systematic reviews that shed light on the latest advancements, challenges, and potential directions in the field of radiology-specific foundation models, appealing to researchers working at the nexus of AI, machine learning, and applied radiology.

Are you interested in contributing a manuscript to this Research Topic?

Multimodal Foundation Models in Radiology